Estimating the abundance of polar bears on Wrangel Island during late summer using high-resolution satellite imagery: a pilot study

Remote sensing techniques for automated marine mammals detection: a review of methods and current challenges

Marine mammals are under pressure from multiple threats, such as global climate change, bycatch, and vessel collisions. In this context, more frequent and spatially extensive surveys for abundance and distribution studies are necessary to inform conservation efforts. Marine mammal surveys have been performed visually from land, ships, and aircraft. These methods can be costly, logistically challenging in remote locations, dangerous to researchers, and disturbing to the animals. The growing use of imagery from satellite and unoccupied aerial systems (UAS) can help address some of these challenges, complementing crewed surveys and allowing for more frequent and evenly distributed surveys, especially for remote locations. However, manual counts in satellite and UAS imagery remain time and labor intensive, but the automation of image analyses offers promising solutions. Here, we reviewed the literature for automated methods applied to detect marine mammals in satellite and UAS imagery. The performance of studies is quantitatively compared with metrics that evaluate false positives and false negatives from automated detection against manual counts of animals, which allows for a better assessment of the impact of miscounts in conservation contexts. In general, methods that relied solely on statistical differences in the spectral responses of animals and their surroundings performed worse than studies that used convolutional neural networks (CNN). Despite mixed results, CNN showed promise, and its use and evaluation should continue. Overall, while automation can reduce time and labor, more research is needed to improve the accuracy of automated counts. With the current state of knowledge, it is best to use semi-automated approaches that involve user revision of the output. These approaches currently enable the best tradeoff between time effort and detection accuracy. Based on our analysis, we identified thermal infrared UAS imagery as a future research avenue for marine mammal detection and also recommend the further exploration of object-based image analysis (OBIA). Our analysis also showed that past studies have focused on the automated detection of baleen whales and pinnipeds and that there is a gap in studies looking at toothed whales, polar bears, sirenians, and mustelids.

Marine mammal conservation: over the horizon

Marine mammals can play important ecological roles in aquatic ecosystems, and their presence can be key to community structure and function. Consequently, marine mammals are often considered indicators of ecosystem health and flagship species. Yet, historical population declines caused by exploitation, and additional current threats, such as climate change, fisheries bycatch, pollution and maritime development, continue to impact many marine mammal species, and at least 25% are classified as threatened (Critically Endangered, Endangered or Vulnerable) on the IUCN Red List. Conversely, some species have experienced population increases/recoveries in recent decades, reflecting management interventions, and are heralded as conservation successes. To continue these successes and reverse the downward trajectories of at-risk species, it is necessary to evaluate the threats faced by marine mammals and the conservation mechanisms available to address them. Additionally, there is a need to identify evidence-based priorities of both research and conservation needs across a range of settings and taxa. To that effect we: (1) outline the key threats to marine mammals and their impacts, identify the associated knowledge gaps and recommend actions needed; (2) discuss the merits and downfalls of established and emerging conservation mechanisms; (3) outline the application of research and monitoring techniques; and (4) highlight particular taxa/populations that are in urgent need of focus.

A comparison of baleen whale density estimates derived from overlapping satellite imagery and a shipborne survey

As whales recover from commercial exploitation, they are increasing in abundance in habitats that they have been absent from for decades. However, studying the recovery and habitat use patterns of whales, particularly in remote and inaccessible regions, frequently poses logistical and economic challenges. Here we trial a new approach for measuring whale density in a remote area, using Very-High-Resolution WorldView-3 satellite imagery. This approach has capacity to provide sightings data to complement and assist traditional sightings surveys. We compare at-sea whale density estimates to estimates derived from satellite imagery collected at a similar time, and use suction-cup archival logger data to make an adjustment for surface availability. We demonstrate that satellite imagery can provide useful data on whale occurrence and density. Densities, when unadjusted for surface availability are shown to be considerably lower than those estimated by the ship survey. However, adjusted for surface availability and weather conditions (0.13 whales per km², CV = 0.38), they fall within an order of magnitude of those derived by traditional line-transect estimates (0.33 whales per km², CV = 0.09). Satellite surveys represent an exciting development for high-resolution image-based cetacean observation at sea, particularly in inaccessible regions, presenting opportunities for ongoing and future research.

Using Deep Learning to Count Albatrosses from Space: Assessing Results in Light of Ground Truth Uncertainty

Many wildlife species inhabit inaccessible environments, limiting researchers ability to conduct essential population surveys. Recently, very high resolution (sub-metre) satellite imagery has enabled remote monitoring of certain species directly from space; however, manual analysis of the imagery is time-consuming, expensive and subjective. State-of-the-art deep learning approaches can automate this process; however, often image datasets are small, and uncertainty in ground truth labels can affect supervised training schemes and the interpretation of errors. In this paper, we investigate these challenges by conducting both manual and automated counts of nesting Wandering Albatrosses on four separate islands, captured by the 31 cm resolution WorldView-3 sensor. We collect counts from six observers, and train a convolutional neural network (U-Net) using leave-one-island-out cross-validation and different combinations of ground truth labels. We show that (1) interobserver variation in manual counts is significant and differs between the four islands, (2) the small dataset can limit the networks ability to generalise to unseen imagery and (3) the choice of ground truth labels can have a significant impact on our assessment of network performance. Our final results show the network detects albatrosses as accurately as human observers for two of the islands, while in the other two misclassifications are largely caused by the presence of noise, cloud cover and habitat, which was not present in the training dataset. While the results show promise, we stress the importance of considering these factors for any study where data is limited and observer confidence is variable.

A comparison of drone imagery and ground-based methods for estimating the extent of habitat destruction by lesser snow geese (Anser caerulescens caerulescens) in La Pérouse Bay

Lesser snow goose (Anser caerulescens caerulescens) populations have dramatically altered vegetation communities through increased foraging pressure. In remote regions, regular habitat assessments are logistically challenging and time consuming. Drones are increasingly being used by ecologists to conduct habitat assessments, but reliance on georeferenced data as ground truth may not always be feasible. We estimated goose habitat degradation using photointerpretation of drone imagery and compared estimates to those made with ground-based linear transects. In July 2016, we surveyed five study plots in La Pérouse Bay, Manitoba, to evaluate the effectiveness of a fixed-wing drone with simple Red Green Blue (RGB) imagery for evaluating habitat degradation by snow geese. Ground-based land cover data was collected and grouped into barren, shrub, or non-shrub categories. We compared estimates between ground-based transects and those made from unsupervised classification of drone imagery collected at altitudes of 75, 100, and 120 m above ground level (ground sampling distances of 2.4, 3.2, and 3.8 cm respectively). We found large time savings during the data collection step of drone surveys, but these savings were ultimately lost during imagery processing. Based on photointerpretation, overall accuracy of drone imagery was generally high (88.8% to 92.0%) and Kappa coefficients were similar to previously published habitat assessments from drone imagery. Mixed model estimates indicated 75m drone imagery overestimated barren (F2,182 = 100.03, P < 0.0001) and shrub classes (F2,182 = 160.16, P < 0.0001) compared to ground estimates. Inconspicuous graminoid and forb species (non-shrubs) were difficult to detect from drone imagery and were underestimated compared to ground-based transects (F2,182 = 843.77, P < 0.0001). Our findings corroborate previous findings, and that simple RGB imagery is useful for evaluating broad scale goose damage, and may play an important role in measuring habitat destruction by geese and other agents of environmental change.